In Ancient Quenya, any surviving medial voiced stops usually became voiced spirants, the most notable exception being those appearing after nasals. This sound change also excludes voiced stops that underwent other earlier changes, such as when voiced stops became nasals before nasals. Tolkien described the spirantalization of voiced stops in numerous places (PE17/127, 135, 153; PE19/18, 23; PE22/44, 149). The most detailed descriptions appeared in the Outline of Phonetic Development (OP1) from the 1940s and the Outline of Phonology (OP2) from the 1950s:

The voiced stops b, d, g. These all became weakened, ceasing to be stops and becoming continuants or weak spirants. b became ƀ (bilabial v) ... d became l initially. Medially between vowels it would appear to have been slightly retracted and to have become a weak untrilled [r], distinct from original r ... g probably became [ʒ] but this stage was already past in early PQ (OP1: PE19/32).

The voiced stops. These all became weakened, ceasing to be stops and becoming (weak) spirants or continuants. b became bilabial ƀ ... d became l initially. Medially between vowels it appears to have been somewhat retracted, and to have become at first a weak untrilled r or đ-like sound distinct from original r. The two sounds have distinct letters in earlier Quenya spelling, which may be represented by r (original) and r̃ (derived from d). 7 = r, 6 = ř ... g became at first the open spirant ʒ, with weak friction (OP2: PE19/69-70).

The resulting spirantal sounds then underwent further developments:

• Bilabial ƀ [β] became labio-dental v.
• The result of medial d generally became r.
• The voiced velar spirant ʒ [ɣ] usually vanished.

The medial and initial developments of d were somewhat different. As noted above, initial [d] became [l], but the medial sound ultimately became [r]. Tolkien isn’t entirely clear on exactly how the d developed medially: he described the result as an “weak untrilled r or đ-like sound”. Thus, [d] might have become either [ð] or [ɹ] as in the English pronunciation of “r”, but more likely it passed through both on its way to becoming [r]. For simplicity and consistency with other voiced stop spirantalizations, the Eldamo data model represents this stage of its phonetic development as [d] > [ð], and its passage through [ɹ] is discussed in the entry on how intervocalic [ð] became [r]. This full [d] > [ð] > [ɹ] > [r] phonetic development was described by Tolkien in notes on the history of ar “and”, probably from the mid-1960s:

Ad went through four stages of development: (1) archaic before records ad (2) early weakening ađ (3) retraction to ař, an untrilled spirantal r (as commonly heard in English “dry”), (4) assimilation to normal r (PE17/71).

The conditions under which stops were preserved rather than spirantalized evolved over time. In the 1930s, the voiced stops were preserved after both nasals and liquids: “Owing to the opening in Q. of all voiced stops (except after nasals and l, r) ... (OP1: PE19/40)”. In the 1950s, Tolkien reaffirmed that voiced stops were preserved after nasals: “[Voiced stops] after nasals. In this position the stops remained as such, and mb, nd, ñg were among the most favoured combinations in Quenya (OP2: PE19/92).” However, in OP2 it seems they first became spirantal between a liquid and a vowel, and only afterwards did they (mostly) became stops again:

[Voiced stops] after r, l. See above under ii (g), for a discussion of the processes of change here involved. It is probable that in AQ the voiced stops were at first opened to spirants after r, l; but that a reaction soon supervened, while ƀ, ð, ʒ were still at that stage, and identical in place of articulation with the original stops (as surviving in mb, nd, ñg(w)), causing some at least of these spirants to be re-stopped.

The net result is that voiced stops only appear after nasals and liquids in modern spoken Quenya (Tarquesta), as it was conceived in both the 1930s and 1950s; it is just that in the 1950s the development after liquids was stops > spirants > back to stops. See the discussion of how spirants became stops after [r], [l] for more details. In both conceptual periods, voiced stops were also preserved in three-consonant clusters like lgy, rgy (PE19/40, 93); see below for discussion.

In the 1930s, spirantalization occurred before velars were dentalized before [j]:

[Initial] gy became a spirantal y, but this fell in, in PQ, with original y (OP1: PE19/34).

k, g were fronted to t, d before y. Owing to the opening in Q. of all voiced stops (except after nasals and l, r) gy > dy only occurred in ndy, ldy, rdy (< ñgy, lgy, rgy) (OP1: PE19/40).

As indicated by this note, voiced stops did survive in three-consonant clusters, so that such gy combinations were ultimately dentalized to dy; the same was true in the 1950s (PE19/83, 93). In OP2, however, spirantalization was different from the 1930s in that it occurred in two phases: medial stops became spirants before dentalization, but initial stops were spirantalized after dentalization:

The dentalization of k-series before (initial) y was early: it is also shared by Telerin. Thus gy > dy or đy ... Medial gy became ı̯y, which coalesced with preceding vowels or formed diphthongs: thus CE magyā > maiya, maia. This was because opening of the medial stops was much earlier than in case of the initials; so that gy already to ʒy > ı̯y before dentalization (OP2: PE19/75).

This splitting of spirantalization into two phases first appeared in revisions to OP2 (PE19/75 note #37) and almost certainly coincides with the introduction of the new initial cluster ly into Quenya (PE19/80 note #54). See the discussion on how [dj] became [lj] for more details.

Conceptual Development: In the Early Qenya of the 1910s and 1920s, medial voiced stops became voiced spirants with phonetic developments similar to Tolkien’s later ideas: b, d, g > ƀ, ð, ʒ > v, r, nil. Roots from the Qenya Lexicon used the marker Ř to indicate [r] that might have been originally derived from primitive medial D. This analysis is complicated by the fact that voiced spirants were also part of the phonetic inventory of Primitive Elvish in the earliest conceptual period, so that some of these Ř might represent primitive [ð] instead, which Tolkien sometimes specified in roots as Đ.

In this early period, the initial development of voiced stops was different from the sound changes in the 1930s and 50s. In the 1910s and 20s, initial voiced stops were unvoiced to p, t, k, as in:

• ᴱ✶balga > ᴱQ. palla “paunch” vs. ᴱN. bala “hump” (PE13/138).
• ᴱ✶dagla > ᴱQ. taila vs. ᴱN. dail “axe (blade)” (PE13/141; PE14/66).
• ᴱ√GOÞO [ᴱ√KOSO in the Qenya Lexicon] > ᴱQ. kos “strife” vs. G. goth (QL/48; GL/42; PE13/105).

Tolkien probably revised these initial developments at the same time he introduced his new paradigm for Primitive Elvish in the 1930s. Among other things, he removed voiced spirants from its phonetic inventory.

Neo-Quenya: As discussed under the entry on how [dj] became [lj], I recommend ignoring Tolkien’s split in the 1950s of spirantalization into two phases and simply assume it occurred before the dentalization in all cases. This allows the preservation of a number of earlier words that show initial gy- > ʒy- > y-, as well as giving a more straightforward explanation for the use of the tengwa hÍ (anna with palatalization tehta) for y, since this combination originally represented ʒy. Furthermore, there is only one attested Quenya word that begins with ly: lye “(polite) you” which arose by analogy with -n(yë) rather than derivation from gy-. However, if you want to mirror Tolkien’s later conception of these phonetic developments, than earlier forms like ᴹQ. yello and ᴹQ. yerna that are derived from primitive gy- should be adjusted to begin with ly-.